Professional Documents
Culture Documents
1
Abstract
In this paper we will describe the process behind computing traveling-
wave solutions in a diffusive predator-prey system. There is informa-
tion on the existence of traveling-wave solutions using an abstract
topological argument by Dunbar [D] and there are methods for com-
puting traveling-wave solutions by Doedel & Friedman [DF]. But the
difference here is that we shall describe how to actually compute these
solutions so that they can be viewed and further examined. The main
purpose of this paper is to be less theoretical and more practical. If
one would like to view the theory behind the methods used here to
find the traveling-wave solutions, read Beyn [B].
1 Introduction
Mathematical modeling will always be an important field of mathematics
because of its applications to the real world. While no model is perfect, if
a close enough approximation can be obtained, then scientists can see how
certain factors will affect a situation by merely working out equations on a
piece of paper as opposed to actually running an experiment.
2
state. That is, initially there would be a small amount of turtles at the one
end, but as they grow in numbers from eating the plants, eventually they will
spread out. If one were looking at this process on a space - population graph
that changed over time, it would look as if a wave of turtles was moving
across the coastline.
Scientists that have studied real pred-prey systems for certain species such
as plankton and also in reaction-diffusion equations from neurophysiology,
chemistry and epidemiology have seen that traveling waves indeed exist in
their respective areas of study. Since traveling waves exist in practice, it is
essential that they are represented in the mathematical model. Traveling
waves are a special, fundamental class of solutions of PDEs used to represent
the transportation of information in a single direction. The ascertainment
of the traveling waves was a major discovery in the pred-prey model [M]
because of how it represented a transition between equilibria.
du
= au buw, (1)
dt
dw
= ew + f uw, (2)
dt
where a, b, e, f are positive constants. As is apparent from looking at the
model, the prey shall grow exponentially depending on the difference between
rate of growth, a, and the amount of prey being eaten, bu. The predators
shall grow exponentially depending on the difference between the amount of
prey being eaten, f u, and the rate of predator death without prey, e. We
note that if w = 0, then u grows exponentially without bound.
3
As can be seen in figure 1, there are closed orbits surrounding the critical
point ( ab , fe ) that implies the solutions are periodic in time and that the
maxima of each population are shifted from each other. There is a saddle
point at (0, 0) representing the absence of both populations. The saddle
point shows what is obvious in real life, if there are prey and no predators,
the prey will grow. If there are predators and no prey to eat, then they will
die out because of the absence of food.
x=x1xy
y=1y+xy
2.5
1.5
y
0.5
du u
= au(1 ) buw, (3)
dt K
dw
= ew + f uw, (4)
dt
where K is the maximum amount of prey that can exist in the model. There-
fore if K = 5, then if a small amount of prey was added to a system with no
predators, the prey would grow to 5 as t as apposed to the prey grow-
ing to as t . From looking at (3), if the prey population is greater
4
u
than the carrying capacity, the (1 K
) term will be negative ensuring that
the prey population will decrease.
Now by examining figure 2, there is a saddle point at (K, 0), showing that
if any solution starts with more than K prey, then they will quickly go back
to the < K region. Also, now instead of there being a spiral point, there is
a stable point that all solutions approach as t .
There are 3 critical points of the system. One is (0, 0) which is the absence
of both species. The second critical point, (K, 0), represents the situation
where the prey have reached their carrying capacity in the absence of preda-
tors. The final critical point is ( fe , ab (1 feK ), which is the coexistence state
where both species are nonzero. The first two critical points are unstable
saddle points and the coexistence critical point is a stable critical point that
is approached as t .
e a e
u(+) = , w(+) = (1 ).
f b fK
x = 2 x (1 x/2) 1 x y
y=1y+xy
2.5
1.5
y
0.5
5
some sort of distribution over space (for simplicity we shall analyze one-
dimensional space).
Once the concept of space has been added to the equations, another
concept must be examined, which is that populations will move between
space. For example, if a large amount of some population is crammed into
a certain area, they will tend to spread out to maximize productivity. This
process is called diffusion.
c J
= ,
t x
c
(D x )
= ,
x
2c
= D 2.
x
Now inserting the diffusion term into the pred-prey equations we have func-
tions for the pred and prey in terms of space and time
u
ut = D1 uxx + au(1 ) buw, (5)
K
wt = D2 wxx ew + f uw, (6)
6
what relationships between the terms are important. By a straightforward
calculation we get
Ut = DUxx + U (1 U W ), (7)
Wt = Wxx + W (U ), (8)
where
u b x
U= , W = w, t0 = at, x0 = q ,
K a D2 /a
D1 fK e
D= , = , = ,
D2 a fK
and we have dropped the 0 on t and x for readability.
The next step is to assume that the solutions of the diffusive model (7)
(8) have the shape of a traveling wave as it propagates. While there could be
other types of solutions that are not traveling waves, we are only interested
in the traveling-wave solutions. Therefore we shall make the assumption that
there exists some solutions that have the following form
where c is the wave speed. Now one problem with the model is that it is
a PDE which is much more difficult to deal with. To simplify things let
s = x + ct (s is called the wave variable). After applying the chain rule, we
can get a system in terms of just s. The model becomes
7
where 0 = d
ds
.
u0 = v, (11)
c 1
v0 = v u(1 u w), (12)
D D
w0 = z, (13)
z 0 = cz w(u ). (14)
Now there is a first-order system of equations to try and find a traveling-wave
solution for. It should satisfy the following boundary conditions:
u() = 1, w() = 0,
u(+) = , w(+) = 1 .
Finding a traveling-wave solution of the PDE (7)(8) is equivalent to find-
ing a trajectory in the four-dimensional phase space that satisfies the above
boundary conditions.
8
3.1.1 Critical point (1, 0, 0, 0)
The Jacobian Matrix of the functions on the RHS of the ODE (11)(14) is
0 1 0 0
2 1 1 c 1
D
u + D
w D D D
u 0
.
0 0 0 1
w 0 u c
The eigenvalues of the function on the RHS of the ODE (11)(14) at (1, 0,
0, 0) are
1 = (c c2 + 4D)/2D,
2 = (c + c2 + 4D)/2D,
q
3 = (c c2 4(1 ))/2,
q
4 = (c + c2 4(1 ))/2.
By analyzing these eigenvalues, certain restraints on parameters are ob-
tained. First of all, D 1. This will ensure that the exiting solutions do
not fall into the w region. Since we are looking at a biologically interesting
model, we do not want to deal with the situation of having negative amount
of predators.
q
With the same goals in mind, there is the restraint c > c = 4(1 ).
If this were not the case, then there would be 2 complex conjugate eigenval-
ues. This would cause a spiral around (1, 0, 0, 0) which would lead to w < 0
as s . Figure 3 shows this scenario with c < c . The solutions oscillate
and that forces the amount of predators to be less than 0.
The eigenvalues at (1, 0, 0, 0), are all greater than 0 except for 1 .
Therefore there is a 3-dimensional unstable manifold at (1, 0, 0, 0).
Let
9
1.2
0.8
0.6
0.4
0.2
0.2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
p(0 , 1) = (1 ) 2 /4,
therefore if < = 4(1) then p(, 1) will have two negative real roots.
Figure 4 shows the scenario of what will happen if > . The critical
point (, 1 ) is approached with damped oscillations because there are
2 imaginary stable eigenvalues. If < , then the critical point would be
approached monotonely. For a more robust discussion on analyzing the value
of refer to Dunbar [D].
Also using the Hurwitz algorithm on p(), we see that there will always be
2 negative real part eigenvalues and 2 positive real part eigenvalues. There-
fore there is a 2-dimensional unstable manifold and a 2-dimensional stable
manifold at (, 0, 1 , 0).
10
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Another problem is that the solution that connects the two points is
very unstable. Let x be the actual solution with x , x+ as the stationary
points (1, 0, 0, 0) and (, 0, 1 , 0) respectively. Let nu be the number of
dimensions for the unstable manifold at x , n+s be the number of dimensions
for the stable manifold at x+ and n be the total number of dimensions in
the system of equations. If nu + n+s = n + 1 (which it is in our case, 3
+ 2 = 4 + 1), then there is only one solution. Combined with the limited
amount of precision available on computers, using methods such as shooting
and estimating points that should be on the solution are ineffective.
11
the solution should end. But with a BVP solver, one can pass a certain
amount of conditions (that being the number of dimensions the system of
equations is) about the starting point and the end point. It is possible to
pass in x(0)u = 1, x(1)w = and also relations between points such as
x(0)v + x(1)z = 0.
Let y be defined as the first section of x such that y(0) = x(0) and z
be defined as the next section of x such that z(1) = x(1). Then naturally
y(1) = z(0). This will create 4 boundary conditions
y(1) z(0) = 0.
xTd (y(1) xm ) = 0.
12
unstable eigenvectors at x+ . Then define Ls as a left stable eigenvector at
x and L+u as a left unstable eigenvector at x+ such that
Ls vs = 1,
L+u v+u = 1.
The left eigenvalues can be found as row i of A1 where i is the corresponding
eigenvector column in A.
Ls (y(0) x ) = 0,
L+u (z(1) x+ ) = 0.
4 Conclusion
The goal of my research was to discover a way to compute traveling-wave
solutions in the diffusive Lotka-Volterra pred-prey model and that is what I
have attained. Many mathematical papers discuss traveling waves and the
diffusive pred-prey model, yet they tend to be more theoretical. In my paper
I have shown exactly how to view these solutions for this model by using
Matlab (some papers discuss using AUTO to find traveling-wave solutions,
but using Matlab is much more simple). This gives the opportunity to see
exactly how changing the parameters affect the model. Although we have
shown generally how affecting the parameters would affect the model, now
one could get precise numbers on the effects.
13
4.1 Future Work
There are two pretty significant problems with the logistic growth model
that was used in my paper. First of all there is a stable critical point that is
approached as t , which does not make much sense because it is unlikely
that the predator and prey populations will stay at some certain amount
Also there should be some maximum amount of prey that the preda-
tors can eat. Both of these problems are resolved by adding another factor
called the satiation effect. This is a more complicated model that could be
examined.
Finally in the real world, the prey would tend to run away from the
predators and the predators would in turn chase after the prey. This is
another diffusion term that could be added to the model and analyzed.
4.2 Acknowledgements
To Gregory Lyng for his help and dedication throughout this project.
The code used to create the phase planes (figure 1 and 2) was written by
John C. Polking, Rice University.
References
[M] Murray, J.D., Mathematical Biology, Springer-Verlag, 1989.
14
[B] Beyn, W.J., The Numerical Computation of Connecting Orbits in Dy-
namical Systems, IMA Journal of Numerical Analysis, 9, 1990, 379-405
Appendix
A Matlab Traveling-Wave Solutions Code
function ppTravelingWaves
% append solutions
t = [eta (ones(1,size(eta))+eta)];
u = [f(1,:) f(5,:)];
v = [f(2,:) f(6,:)];
w = [f(3,:) f(7,:)];
z = [f(4,:) f(8,:)];
15
% plot solutions
plot(t, pp);
%-------------------------------------------------------------------------
u = f(1);
v = f(2);
w = f(3);
z = f(4);
u2 = f(5);
v2 = f(6);
w2 = f(7);
z2 = f(8);
deriv = [u_prime;
v_prime;
w_prime;
16
z_prime;
u2_prime;
v2_prime;
w2_prime;
z2_prime];
%--------------------------------------------------------------------------
u0 = 1;
w0 = 0;
u1 = b;
w1 = 1-b;
% compute eigenvectors
[E0, D0] = eig(A0);
[E1, D1] = eig(A1);
17
if D0(1,1) < 0;
L0s = E0inv(1,:);
else
disp(error: first eigenvalue is unstable);
end
18
(yb(4) - ya(8))
p0
p1
p2
phase];
19